OpenAI’s API has revealed three new models: aurora-green-alpha aurora-blue-alpha aurora-red-alpha About an hour later, three new models appeared on Design Arena: aspen, cypress, and maple. This strongly suggests OpenAI is testing new image generation models.

Claude Code just added idle compaction, and this could save a LOT of usage. Imagine you have a session with 500K tokens of context. You come back 90 minutes later, your prompt cache has expired, and Claude has to process all 500K tokens again without the cache. That's insanely expensive. Now, Claude automatically compacts idle sessions before the cache expires. Instead of reprocessing 500K tokens, you might only need to process, say, 100K. And imagine the difference with an 800K-token session. I always recommend manually compacting after finishing parts of a task, but most people never do it. This feature could save the average Claude Code user a ton of usage.

We recently added idle compaction to Claude Code! It now compacts long idle sessions right before the prompt cache expires (~1h). If you come back to it later, it only has to reprocess the summary instead of the entire conversation, which saves you some usage.

Wondering how he keeps catching AI model leaks before everyone else? He literally built his own AI model radar. And the best part? He OPEN-SOURCED it. You can run it yourself star and fork his repo: 🔗 Or join his Discord and get the alerts without setting up anything:

Gemini 4 Argon benchmarks are out We've actually seen these benchmarks before, back when the model was first announced. Looks really strong in text, vision, etc. thanks to Google's foundation. But I feel like the coding scores are heavily benchmark-optimized. Happy to be wrong.

Gemini 4 Argon is here. It’s available globally today for developers and enterprises in the paid tier of the Gemini API and the Gemini platform, for Ultra subscribers in Google Antigravity, and Ultra and Pro subscribers in the Gemini app and Search. Hope you all enjoy it!

Day 8 of the OpenAI sprint: I feel like OpenAI just doesn’t give a damn about its users. For the third day in a row, there’s been no news before 10 PM UTC, when most of the world is already on Tuesday. I think the whole sprint idea has failed, and they’re just afraid to admit it.

Congrats, X Premium subscribers! For $8/month, you can now use Cursor Composer instead of Grok, which burns through its limits way too fast. And X Premium+ subscribers, your subscription now costs $30/month and includes Grok Bot!

Week 1 of the OpenAI sprint: recap. Day 1: WIN Speed boost from ~30 to ~50 TPS for Sol and Astra. Most people use these models, so this is a clear win. Day 2: REJECTED Auto-review, API tiers, meeting notes, Decisions API. Honestly, all pretty useless to me. I'm always on bypass permissions anyway. Decisions API should just be included in the subscription, classifiers are dirt cheap. Day 3: WIN GPT-6 finally shipped to ChatGPT. This was genuinely great. Faster responses and web search too. Easily one of the best days. Day 4: REJECTED Tailscale support was apparently already there, just announced as a new feature. Instant steering is decent, but Sol Ultrafast is for the Pro tier. Not enough. Day 5: REJECTED Composer Predictions and Dots updates, both Pro only. These don't meet the criteria of being relevant to most users. Most people aren't on Pro. We need these features on Plus and lower tiers. Day 6: RESET Reset delivered. Fair enough, exactly what was promised. Day 7: REJECTED Default context increased to 500K when 1M was already available manually. No extra long-context fees when subscriptions never had them anyway. Basically fake improvements. Week 1 stats: 2 wins 4 rejected days 1 reset day + 1 bonus reset (on day 3 for Codex reaching 30M active users) The speed boost and GPT-6 in ChatGPT were the clear highlights. Everything else was either too limited, already available, or not really an improvement. Let's see what Week 2 brings.